Mirendil Raises $1B Series A to Automate Frontier AI Research

Mirendil has raised $1 billion in a Series A led by Kleiner Perkins and Clearlake Capital, the company announced on November 12, 2026. The round backs an unusual pitch: making frontier AI research cheap enough for individual scientists and small teams.

Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 participated, alongside other new and existing investors. The company's stated goal is that one person can take on work that previously required a team, and a team can pursue projects that previously required an institution. That is a claim about capability, not yet a demonstrated product, and the company itself frames it as a mission rather than a result.

What did the one-researcher experiment actually show?

The round follows Mirendil's first published model results. In a report dated October 21, 2026, the company said a single researcher working with AI agents pretrained Mirendil-Base, a coding model, from scratch in 3.5 months on less than $10 million in compute. That all-in figure covers research and small-scale experiments, the final training run, agent token spending and data processing.

By comparison, frontier training runs at major labs are widely reported to cost far more and involve large teams; a sub-$10 million, one-person run would compress both budgets and headcount by an order of magnitude or more. Mirendil reports that agents wrote more than 99% of the code, ran experiments and diagnosed failures. The researcher chose what to try and trust, and intervened when needed.

The experiment relied on publicly available code, web pages and public datasets. The company says it purchased or licensed no additional data and used no distillation or synthetic data. Mirendil-Base sits in the 100 billion to 500 billion parameter range. The larger Mirendil-Base-Large and Mirendil-Base-XL models were still in training when the report went out. In plain terms: one audited internal result, not yet an independently replicated one.

What are Fern and Lor?

Mirendil has since post-trained Mirendil-Base into Fern 0.1, its first model. The Fern model page describes a system built for high-volume agentic work: debugging training runs, reading complete agent traces and auditing training data. The architecture, training and serving support a context window of up to 10 million tokens, roughly 23 MB, or about half a million lines of code. That window lets the model hold days of an agent's work in view at once. Access is listed as coming soon.

Lor is the delegation layer. Introduced on October 28, 2026, the cloud platform runs agents that write code, run experiments, provision compute and operate on user-selected machines. An agent can split a job across dozens of agents, with hundreds running across projects. Mirendil says agents can work for days or weeks and unblock themselves when something breaks. Users can follow each agent's progress, stop agents and remove machines.

Lor is available by invitation in private beta, free during the beta, on macOS, Linux and the web. Windows, iOS and Android are listed as coming soon. A free beta is not a business model, and the company has not disclosed pricing or launch timing.

How is the build funded and powered?

Mirendil launched on June 24, 2026 with a founding team of 20 researchers and engineers drawn from Anthropic, xAI, Google DeepMind and OpenAI. At launch it disclosed a $200 million seed round led by Andreessen Horowitz and Kleiner Perkins, with follow-on investment from NVIDIA and others. The new round brings disclosed funding to roughly $1.2 billion within roughly five months of launch.

Compute is secured through a multi-year Google Cloud partnership announced August 6, 2026. The arrangement puts training, inference and AI research workloads on Google Cloud's AI Hypercomputer, drawing on Google Cloud TPUs and full-stack NVIDIA AI infrastructure. It includes managed training clusters co-designed across compute, storage, networking and control planes to support larger research loops.

The Series A ties three pieces together: agent-assisted model development, a platform for delegating research work, and cloud infrastructure for bigger experiments. Whether the internal one-researcher result becomes a repeatable workflow for outside scientists is the central test of the accessibility mission behind the billion dollars.